基于CEEMDAN-LSTM模型的污水处理厂N_(2)O排放预测研究

Research on N_(2)O emission prediction in WWTPs based on CEEMDAN-LSTM

在我国实行“双碳”战略的背景下,污水处理厂N_(2)O排放预测对于污水处理厂的碳中和有积极意义。现有的污水处理厂N_(2)O排放预测研究通常直接采用基于包含噪声的N_(2)O排放量数据进行建模预测,导致模型预测精度不高。采用自适应噪声完备集合经验模态分解-长短期记忆网络(CEEMDAN-LSTM)模型,通过引入CEEMDAN方法缓解数据中噪声对模型的影响以提高模型预测精度,对污水处理厂N_(2)O排放进行预测并在预测验证集上验证模型。与LSTM、门控循环单元(GRU)、人工神经网络(ANN)和支持向量机(SVM)模型相比,CEEMDAN-LSTM预测精度最高,均方误差(MSE)、平均绝对误差(MAE)和平均绝对百分比误差(MAPE)分别为5 497.11、56.55和1.22%,能够更高精度地预测N_(2)O排放量,为污水处理厂采取合适的碳中和策略提供理论支撑。

Against the backdrop of China's implementation of the"dual carbon"strategy,predicting N_(2)O emissions from wastewater treatment plants(WWTPs)has significance for carbon neutrality in WWTPs.The existing studies on N_(2)O emission prediction is usually directly based on modeling and prediction of N_(2)O emission data containing noise,resulting in low prediction accuracy of the model.In this study,a model combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and Long Short-Term Memory(LSTM),referred to as the CEEMDAN-LSTM model,is introduced.The CEEMDAN methodology is introduced to alleviate the impact of noise in the data,enhancing the model's predictive accuracy.The model is applied to forecast N_(2)O emissions from WWTPs and is validated on a test dataset.Compared to models such as LSTM,Gated Recurrent Unit(GRU),Artificial Neural Networks(ANN)and Support Vector Machine(SVM),the CEEMDAN-LSTM model demonstrates superior performance in terms of Mean Squared Error(MSE),Mean Absolute Error(MAE),and Mean Absolute Percentage Error(MAPE),with values of 5497.11,56.55,and 1.22%,respectively.It can more accurately predict N_(2)O emissions,providing theoretical support for wastewater treatment plants to adopt appropriate carbon offset strategies.